Abstract: The main achievements of the national key-research project "Innovative Researches on Chinese Numerical Weather Prediction System", which is conducted in 2001—2005 co-sponsored by Chinese Ministry of Science and Technology and China Meteorological Administration are reviewed. Aiming at the development of Chinese next generation numerical weather prediction (NWP) system which will be able to improve the NWP benefited from the availabilities of more powerful computer facilities and observational data with higher temporal and spatial resolutions, the project focuses on four main issues. They are the development of advanced data assimilation system, the development of unified model dynamic core suitable to different scales, optimization of model physics and parallel computing of NWP model in high performance computer environment. It is expected that the applications of advanced technologies in the new NWP system will result in better performance and higher flexibility for the further upgrading of the system along with the further improvement of computer resources and data availability in the near future.As the main results of this project, a new Global and Regional Assimilation and Prediction System (GRAPES in short) is developed. The main components of the system are the data assimilation system GRAPES-3Dvar, the global NWP model GRAPES-global and the regional NWP model GRAPES-Meso. GRAPES-3DVar is a three dimensional variational assimilation system suitable to both regional and global domains with the capability of assimilating unconventional remote sensing data, such as the radiation data from meteorological satellites. Following GRAPES-3DVar, a four dimensional variational data assimilation system (GRAPES-4DVar) is also developed, GRAPES-Global and GRAPES-Meso are based on the same unified dynamic core with hydrostatic or non-hydrostatic options. A package of model physics plug-compatible with the above model dynamic core is developed. In order to exploit the growing computer resources, parallel computing is one of the main issues in the research and the GRAPES software architecture is organized as a three-level hierarchy to make the system easily transplanted to different computers.The preliminary results of real time pre-operation experiments in a period longer than one year are presented and analyzed briefly too. It shows that the performance of GRAPES is encouraging, but the further improvement is still a critical task to meet the needs of high quality forecasts of severe weather events in China.